Allow joining executions to jobs scoped by state

[segiddins]

To do this, we need arel to be able to track the current table name for jobs (otherwise we get ambiguous column errors), so we construct arel nodes instead of using raw sql strings

This allows me to do something like:

Filter.new({}).filtered_query(state:)
.joins(:executions).then do |rel|
      rel.pluck(
        *rel.group_values,
        Arel::Nodes::Max.new(
          [Arel::Nodes.build_quoted(now, rel.arel_table[:created_at]) -
            rel.arel_table.coalesce(*rel.send(:arel_columns, %w[executions_good_jobs.performed_at executions_good_jobs.finished_at executions_good_jobs.scheduled_at executions_good_jobs.created_at]))]
        )
      )
        .to_h { |*a, v| [a, v] }
    end

[bensheldon]

This is great! Thank you! I just pushed up some fixes for some naming collisions and I'll merge.

[segiddins]

Ah nice I was just about to fix that :)

In case it interests you, this is going towards rubygems/rubygems.org#3582, which adds some statsd instrumentation that is aimed at a more job-centric latency than an individual execution one (basically, we had a few days of retries pile up when we switched one job over to GoodJob and our dashboard just showed the queue depth, but not the fast that some of them shouldve been run days ago)

[bensheldon]

In case it interests you, this is going towards rubygems/rubygems.org#3582, which adds some statsd instrumentation that is aimed at a more job-centric latency than an individual execution one

Yikes! Sorry that the dashboard gave a false sense that everything was fine 😬 I want to improve those charts in #438 (I have some bad SQL in #888). For queue latency I'm thinking the query needs to be something like COALESCE(performed_at, NOW()) - COALESCE(scheduled_at, created_at) because that will allow jobs that haven't begun to run at all to accumulate latency in the calculation by using NOW().

[segiddins]

Ah no it wasn't your dashboard :) it was the one I built off of our statsd metrics, and I knew with was wrong

[bensheldon]

Ah no it wasn't your dashboard :)

Well, my Dashboard also has that problem 😭

This and #885 was released in https://github.com/bensheldon/good_job/releases/tag/v3.14.1

[ollym]

@bensheldon @segiddins hang about, can you make this change also for GoodJob::Execution seems inconsistent to do it for Job and not Execution.

I'm asking as I need to update my patch here:
#876 (comment)

And would be far easier if i can just override the params_execution_count and params_job_class methods in both classes.

[bensheldon]

@ollym oops, sorry about that! I'll make that change today.